Public opinion inversion identification method based on emotion energy dynamic change

Abstract

The invention relates to the technical field of network public opinion monitoring, and discloses an emotion energy dynamic change-based public opinion inversion recognition method, which comprises the following steps of: constructing a weighted emotion potential energy field, and calculating effective emotion potential energy of nodes; and then calculating semantic field topological stress representing the viewpoint structure tearing degree and emotional potential energy field system entropy representing the energy distribution disorder degree, on this basis, constructing an entropy-stress two-phase state space and a dynamic hysteresis loop model, quantifying the social damping characteristics of the system, monitoring the system state in real time, and determining the state of the system. When the two-phase space yield judgment condition and the damping failure judgment condition are met at the same time, a public opinion inversion early warning signal and a risk probability index are output, the critical characteristics of public opinion system instability can be accurately recognized from the dimensions of structural mechanics and dynamics, and the false alarm rate of inversion early warning is effectively reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of network public opinion monitoring, in particular to a public opinion reversal identification method based on dynamic changes of emotional energy. BACKGROUND

[0002] With the popularity of social media, network public opinion evolution presents high frequency, suddenness and nonlinearity. Public opinion reversal, as a special public opinion phenomenon, usually shows a fundamental reversal of public group attitude in a short time, often accompanied by great social influence. Existing public opinion monitoring technology mainly relies on natural language processing and data mining methods, usually adopting keyword matching, topic clustering and sentiment classification methods based on sentiment dictionary or deep learning. These methods mainly focus on statistical analysis of external appearance characteristics of public opinion, such as monitoring discussion heat, number of retransmissions or proportion change of negative sentiment in a specific time period.

[0003] However, relying only on external statistical characteristics cannot accurately understand the internal mechanism and critical state of public opinion evolution. On the one hand, existing monitoring methods usually ignore the topological structure difference of views in semantic space, and cannot effectively quantify the degree of structural tearing of public opinion inside, for example, high heat negative discussion may only be a polarization of single view group, and does not necessarily lead to public opinion reversal; while some heat has not yet broken out public opinion may be in a critical unstable state due to the existence of intense view confrontation stress. The existing technology lacks quantitative means for such deep structure, which easily leads to false alarm or missed alarm of early warning.

[0004] On the other hand, the existing technology lacks modeling of public opinion system dynamics evolution inertia. Public opinion system, as a complex social dynamics system, has a certain buffering and digestion capacity to external information impact. Simply monitoring the instantaneous change of emotional state is easily disturbed by network noise or short-term emotional fluctuations, and it is also difficult to capture the dynamics signs of system damping failure in the silent period before the reversal occurs, which leads to the existing early warning mechanism often lagging behind the actual reversal event, and cannot provide effective advance for management decision. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a public opinion reversal identification method based on dynamic changes of emotional energy, which solves the problem that the existing public opinion monitoring technology only relies on external heat statistics and sentiment proportion analysis, cannot quantify the degree of tearing of public opinion internal view structure, and lacks effective modeling of system dynamics inertia, thereby leading to lagging of reversal early warning and high false alarm rate in complex network environment.

[0006] To achieve the above purpose, the present application is implemented by the following technical scheme: a public opinion reversal identification method based on dynamic changes of emotional energy, comprising the following steps: Construct a weighted emotional potential field: acquire online public opinion data within a set time window, map network nodes to semantic vectors, and calculate the effective emotional potential of each node; Calculate internal state parameters: Based on the weighted emotional potential energy field, calculate the semantic field topological stress that characterizes the degree of tearing in the viewpoint structure, and the emotional potential energy field system entropy that characterizes the degree of disorder in energy distribution. Construct a dynamic response model: calculate the cumulative external stimulus input of online public opinion, and couple the semantic field topological stress with the emotional potential field system entropy into a biphasic state modulus. Generate a hysteresis loop based on the dynamic relationship between the cumulative external stimulus input and the biphasic state modulus. Perform critical instability judgment: monitor the evolution characteristics of the biphase state modulus and the hysteresis loop, and output a public opinion reversal early warning signal when the dual escape judgment is satisfied; The dual escape criterion includes simultaneously satisfying the two-phase space yield criterion and the damping failure criterion.

[0007] The calculation of the effective emotional potential energy of each node includes: The text sentiment intensity modulus, propagation centrality weight, and local semantic redundancy of each node are obtained respectively. The product of the emotional intensity modulus and the propagation centrality weight is used as the gain term; The sum of the local semantic redundancy and the preset smoothing coefficient is used as the damping term; The effective emotional potential energy of the node is obtained by calculating the ratio of the gain term to the damping term.

[0008] The semantic field topological stress, which characterizes the degree of tearing in the viewpoint structure, includes: Identify the mainstream viewpoint clusters in the semantic vector space and calculate the centroid vector of the mainstream viewpoint clusters: Identify dissenting nodes that do not belong to the mainstream view cluster, and calculate the relative position vector between the semantic vector of the dissenting node and the centroid vector; The relative position vector is weighted using the effective emotional potential energy of dissenting nodes, and the equivalent stretching force of dissenting nodes on the mainstream viewpoint cluster is calculated. The modulus of the force is determined as the topological stress of the semantic field.

[0009] The calculation of the entropy of the emotional potential energy field system, which characterizes the degree of disorder in energy distribution, includes: The sum of the effective emotional potential energy of all nodes within the current time window is calculated. Calculate the proportion of the effective emotional potential energy of a single node to the total, and use it as the probability of the node's potential energy proportion. Based on the potential energy proportion probability of all nodes, the entropy of the emotional potential energy field system is calculated using the Shannon entropy definition logic.

[0010] The calculation of cumulative external stimulus input and biphasic state modulus in the construction of the dynamic response model includes: Identify the set of nodes that have newly entered the network within the current time step, and perform time integration or summation on the effective emotional potential of all nodes in the set to obtain the cumulative external stimulus input; The semantic field topological stress and the emotional potential field system entropy are normalized respectively. A two-dimensional state space is constructed with the normalized system entropy as the horizontal axis and the normalized topological stress as the vertical axis. The weighted Euclidean norm of the current state vector in the two-dimensional state space is calculated to obtain the two-phase state modulus.

[0011] The generation of hysteresis loops based on the dynamic relationship between the accumulated external stimulus input and the biphasic state modulus includes: Using the accumulated external stimulus input as the driving variable and the biphasic state modulus as the response variable, an input-response phase trajectory is constructed. Within the observation period, the beginning and end of the phase trajectory are connected to form a closed loop, and the area enclosed by the closed loop is calculated using a numerical integration method. The area of ​​the region is defined as the area of ​​the hysteresis loop, which characterizes the social damping properties of the public opinion system.

[0012] The specific yield determination condition for the two-phase space is as follows: Preset stress yield limit and critical entropy increase threshold; Determine whether the current normalized topological stress is greater than the stress yield limit; Simultaneously determine whether the current normalized system entropy is greater than the critical entropy increase threshold; If both of the above judgment results are yes, then the two-phase space yield judgment condition is satisfied.

[0013] The damping failure determination criteria are as follows: Calculate the moving average of the hysteresis loop area within the historical sliding window as a benchmark reference value; Calculate the ratio of the hysteresis loop area at the current moment to the reference value to obtain the damping attenuation coefficient; calculate the rate of change of the hysteresis loop area over time. If the damping attenuation coefficient is less than the preset area shrinkage threshold and the rate of change is negative, then the damping failure determination condition is met.

[0014] The output of the public opinion reversal early warning signal includes: Perform a Boolean AND operation on the results of the two-phase space yield test condition and the damping failure test condition. Only when the calculation result is true, a reversal warning data packet containing the current biphase state modulus value is generated.

[0015] The method also includes calculating a probability index of public opinion reversal risk: Calculate the deviation of the normalized topological stress, normalized system entropy, and damping attenuation coefficient from their respective preset thresholds. The total deviation is obtained by weighted summation of the various deviations. The total deviation is mapped to a value between 0 and 1 using the Sigmoid function, which serves as the probability index of public opinion reversal risk and is output along with the public opinion reversal warning signal.

[0016] This invention provides a method for identifying public opinion reversals based on dynamic changes in emotional energy. It has the following beneficial effects: 1. This invention constructs a semantic field topological stress calculation model, which utilizes the vector deviation of dissenting nodes relative to the centroid of mainstream viewpoints and the effective emotional potential energy to calculate the stretching force of dissenting groups on the mainstream consensus structure. This method transforms abstract viewpoint conflicts into quantifiable physical stress. Compared with traditional methods based on word frequency statistics or single sentiment classification, it can accurately identify dissenting energy with structural destructive potential, thereby effectively distinguishing between general negative comments and key nodes that may lead to public opinion fragmentation.

[0017] 2. This invention constructs an entropy-stress dual-phase state space, orthogonally coupling the system entropy of the emotional potential energy field, which characterizes the degree of disorder in the system, with the topological stress of the semantic field, which characterizes the degree of structural tearing. When making a reversal judgment, the system state is required to simultaneously satisfy the dual yield conditions of high stress and high entropy, thereby eliminating false signals under a single dimension, such as eliminating polarization states with high heat but stable internal structure, or noise dissipation states with low stress, significantly reducing the false alarm rate of reversal warnings in complex public opinion environments.

[0018] 3. A dynamic hysteresis loop model based on cumulative external stimulus input and biphasic state modulus was established. The social damping characteristics of the public opinion system were quantified by calculating the loop closure area. The system monitors the damping attenuation coefficient in real time and can immediately trigger an early warning at the critical moment when the hysteresis loop collapses and the system's inertial buffering capacity fails. This sudden change in dynamic characteristics often precedes the occurrence of explicit public opinion reversal events, realizing the transformation from post-event detection to pre-event critical early warning. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example: Please see the appendix Figure 1 - Appendix Figure 2 This invention provides a method for identifying public opinion reversals based on dynamic changes in emotional energy, comprising the following steps: S100. Construct a weighted emotional potential field for the public opinion dissemination network: Obtain the public opinion data stream within the target time window, discretize the data nodes, calculate the emotional potential of each node based on the emotional intensity modulus, dissemination centrality weight, and local semantic redundancy, and form a weighted emotional potential field that evolves over time. S200. Calculate the topological stress of the semantic field: Map the public opinion network to the semantic vector space, identify the centroid of the mainstream viewpoint cluster, filter the set of dissenting nodes that have semantic differences from the centroid of the mainstream viewpoint cluster, and calculate the topological stress borne by the system based on the emotional potential energy of the dissenting nodes and their semantic distance from the centroid. Step S300: Calculate the system entropy of the emotional potential field: Statistically calculate the probability distribution of the emotional potential energy of each node in the weighted emotional potential field, and calculate the information entropy that characterizes the orderliness of the system's energy distribution; Step S400: Constructing a hysteresis loop model and dual-space state mapping: Calculate the total potential energy of nodes newly entering the system per unit time as the external stimulus input, construct a dual-phase state space with system entropy and topological stress as coordinate dimensions, establish a dynamic hysteresis loop model between the external stimulus input and the response modulus of the dual-phase state space, and calculate the area change rate of the hysteresis loop. Step S500: Perform critical instability judgment for public opinion reversal: Monitor the area change rate of the hysteresis loop and the evolution trajectory of the two-phase state space. When the preset two-phase space yield condition is met and the area change rate of the hysteresis loop meets the damping failure condition, it is determined that the public opinion system has reversed and an identification signal is output.

[0022] In step S100, the process of constructing the weighted emotional potential field of the public opinion dissemination network first involves the discretization and preprocessing of the original public opinion data stream. This process aims to transform the unstructured continuous text stream into a discrete set of nodes that can be used for physical field operations, and specifically includes the following sub-steps: To collect multi-source public opinion data, the system obtains raw data streams from target social media platforms or news comment sections through application programming interfaces (APIs) or web crawler technology. The collected data fields include at least: Text content, publication timestamps, user unique identifiers, citation or reply relationships, and interaction metadata—the specific implementation techniques for collecting the above data are standard techniques in the field of computer data engineering and will not be elaborated upon here.

[0023] The collected raw data is cleaned and standardized. Regular expressions are used for pattern matching and filtering to address non-semantic noise in the raw text. The specific processing logic includes: HTML tags are removed, URL hyperlinks are filtered, and pre-defined garbled characters and meaningless special symbols are eliminated. Simultaneously, a stop word list is loaded to filter out function words, auxiliary words, and general high-frequency words with no real meaning, retaining nouns, verbs, and adjectives with substantial semantic meaning. The cleaned text data is then marked as valid text. ,in This is the index number of the data node.

[0024] Define public opinion data nodes and their attribute objects, and define each cleaned independent comment or post as a discrete point or node in the public opinion field. The system for each node Instantiate the following property objects: Content attributes: Stores cleaned and valid text. ; Time attribute: Stores the standardized release time. ; Author attribute: Stores the user ID who published this content. This is used for the subsequent calculation of propagation weights; Interactive attributes: Store the number of references or replies to this node at the time of acquisition, as a reference for the initial energy calculation.

[0025] To capture the dynamic characteristics of public opinion evolution, the system executes data slicing based on sliding time windows, and sets the time window length. and sliding step size The continuous time axis is divided into a series of overlapping or non-overlapping computation frames, for any computation time... The system filters out those that meet the requirements. All nodes of the condition This constitutes the set of nodes within the current time window. The subsequent construction of the emotional potential field, calculation of topological stress, and calculation of system entropy are all based on this set. Perform this at the nodes by adjusting the sliding step size. The size of the window can control the temporal resolution of the system's monitoring of public opinion changes; by adjusting the window length... The size of the time window can control the length of the system's memory of historical information and its ability to smooth instantaneous noise. In this embodiment, the advancement of the time window adopts a discrete update method with a fixed step size to ensure the real-time performance and continuity of the calculation process.

[0026] After discretizing and preprocessing the data nodes, the system targets the node set. Each node in Perform the quantification and extraction of attribute parameters. This operation mainly includes the calculation of the sentiment intensity modulus and the calculation of the propagation centrality weight, which is implemented through the following sub-steps: Compute the emotional intensity modulus of the node The system calls a pre-trained natural language processing model to analyze the node content attributes. Semantic analysis is performed. In this embodiment, the BERT model based on the Transformer architecture is used as the sentiment classifier. Specifically, during processing, the... The text is converted into a sequence of word vectors and input into the BERT model. The model then passes through fully connected layers and a softmax layer to output the predefined sentiment category of the text. posterior probability distribution ,in This includes both positive and negative emotions.

[0027] Emotional intensity modulus This is defined as the degree of confidence the model has in determining the sentiment of the text, and its calculation formula is as follows: ; in: Representing text The probability value of being judged as a positive sentiment; Representing text The probability value of being judged as having a negative emotion.

[0028] The range of values ​​for this modulus is: The larger the value, the more vivid and intense the emotion carried by the node. For nodes that the model determines to be neutral or whose modulus value is lower than the preset filtering threshold, the system regards them as invalid emotion nodes and removes them or assigns them very low weights in subsequent calculations. As for the specific network layer number, attention head number and other hyperparameter settings and training process of the BERT model, those skilled in the art can configure them according to the specific computing resources, and will not be elaborated here.

[0029] Compute the propagation centrality weight of the node This step aims to quantify the release nodes. To assess the structural importance of information source users in the public opinion dissemination network and characterize the dissemination penetration of information sources, the system first traverses the data within the current time window and historical backtracking window, extracts the citation, reply, and forwarding relationships between users, and constructs a directed weighted network graph. Among them, vertex set Sideset represents the independent users participating in the discussion. Representing user interaction, the direction of the edge is set from the responder to the respondent, or from the forwarder to the original author, to reflect the flow of influence.

[0030] Based on the constructed network graph The system uses the PageRank algorithm to iteratively calculate the value of each user node. Structural weights The iterative calculation formula is as follows: ; in: :user The PageRank value, i.e., the propagation centrality weight; The total number of users in the network diagram; Damping coefficient, with a range of values ​​of [value missing]. In this embodiment, the value is 0.85, which is used to simulate the probability that the user will continue to click on the link during the browsing process; : Points to user The set of predecessor users, that is, all users who replied or forwarded. User list of content; :user The number of out-degrees, i.e., users The total number of edges pointing to other users.

[0031] The system performs the above iterative calculations until all users in the network have completed them. Value convergence is achieved when the sum of the differences between two consecutive iterations is less than a preset error threshold. After the calculation is completed, the system queries the node. Corresponding author attributes To converge the author Value assigned to node As the propagation centrality weight of this node If the node author is not in the network graph For new users appearing in the system, the system assigns them a preset initial basic weight value. This weight parameter makes the views published by high credibility or core dissemination nodes have a greater energy amplification effect in the emotional potential field.

[0032] The system performs noise suppression processing based on semantic redundancy. This process aims to identify and quantify homogeneous information existing within the current time window. By calculating the local semantic redundancy parameter, it reduces the interference of high-frequency repetitive content on the emotional potential field. Specifically, this is achieved through the following sub-steps.

[0033] The system generates SimHash semantic fingerprints for node text and performs node analysis. text content The system performs word segmentation to extract a set of feature words. For each feature word, the system calculates its hash value and weight value. In this embodiment, the weight value is calculated using the TF-IDF algorithm to reflect the importance of the feature word in the corpus.

[0034] The process of generating a SimHash fingerprint includes: Initialize the dimension as vector (usually taken) ), initialize all components to 0; Iterate through the feature word set, and for each feature word, it... Each bit of the binary hash value is combined with the weight of the feature word. If the hash value's weight is... If the bit is 1, then the vector The Add the weight value to each component; if it is 0, subtract the weight value. After traversal, for the vector Perform dimensionality reduction operation if the vector The If the first component is greater than 0, then the final fingerprint will be... The bit is set to 1 if it is not set to 0 otherwise. Through the above process, the system maps variable-length text content to a fixed-length 64-bit binary integer, i.e., the SimHash semantic fingerprint. .

[0035] The system calculates the Hamming distance between nodes and sets the current node as the reference point. semantic fingerprint With time window All other nodes semantic fingerprint A comparison was made one by one to determine the Hamming distance. Defined as the total number of 1s in the result of an XOR operation between two binary fingerprints, this distance value directly reflects the degree of difference in semantic structure between the two node texts. The smaller the distance, the more similar the text content.

[0036] Local semantic redundancy of computing nodes The system sets a similarity threshold. When Hamming was far away At that time, determine the node For nodes Homogeneous redundant nodes.

[0037] Local semantic redundancy The calculation formula is as follows: ; in: The set of valid nodes within the current time window; : are nodes respectively and nodes SimHash semantic fingerprint; Hamming distance threshold is used to determine whether semantics are substantially the same. : Indicator function, takes the value 1 when the condition in parentheses is met, otherwise takes the value 0.

[0038] The output of this step The significance of numerical physics lies in quantifying the relationship between nodes in the current public opinion field. Information density that is essentially repetitive, if A large value indicates that the viewpoint was repeatedly published in a very short period of time, constituting redundant noise with low information entropy. In subsequent potential energy calculations, this parameter, as a denominator, can automatically reduce the numerical weight of the artificially inflated volume generated by homogeneous spamming behavior, thereby ensuring that the emotional potential energy field reflects a true and effective distribution of public opinion energy, rather than a simple accumulation of invalid data. The specific coding implementation of the SimHash algorithm is a standard technique in the field of computer hash algorithms and will not be detailed here.

[0039] Having obtained the aforementioned emotional intensity modulus Propagation centrality weight and local semantic redundancy Then, the calculations for each node within the current time window are performed. Effective emotional potential within .

[0040] This step is the core of constructing a weighted emotional potential field, which aims to define physical quantities to characterize the actual driving force of a single node on the public opinion dynamic system. Unlike the statistical counting method based solely on emotional polarity in traditional public opinion analysis, the effective emotional potential model proposed in this invention considers the intensity of information, the structural location of the information source, and the scarcity of the information itself.

[0041] The system calculates nodes based on the following mathematical model. Effective emotional potential : ; The specific physical meanings and parameter settings of each symbol in the formula are as follows: :node In the time window The effective emotional potential value within the node. The higher this value, the greater the system's work potential represented by the public opinion information, and the easier it is to trigger a change in the system's state.

[0042] :node The emotional intensity modulus, with a value range of This value, which serves as the basis for potential energy calculation, reflects the intensity of the emotions expressed in the text itself.

[0043] :node The propagation centrality weight, which serves as a gain coefficient, reflects the positional advantage of a node in the social network topology. The opinions published by core nodes with high centrality, even if the emotional intensity is the same, will have their potential impact on the system significantly amplified by this weight.

[0044] :node The local semantic redundancy, which is located in the denominator, plays a role in damping attenuation. When a certain viewpoint is repeatedly published, the marginal potential energy contribution of a single piece of information will decrease sharply according to an inverse proportional function. This mechanism ensures from a mathematical model that only unique viewpoints or first-release viewpoints with high information entropy can maintain high potential energy. Although there are many mechanically repeated online information, the potential energy of a single piece of information approaches zero, thus failing to accumulate enough total energy to change the state of the system.

[0045] The numerical smoothing factor is set to a very small positive floating-point number. Its purpose is to prevent the smoothing effect of numerical smoothing when... The denominator being zero at any time leads to computational overflow, while ensuring the stability of potential energy calculation in the absence of redundancy.

[0046] After constructing the weighted sentiment potential field in step S100, the system executes step S200, which calculates the semantic field topological stress. This step aims to quantify the tearing tension exerted by dissenting opinions on the mainstream consensus structure in the public opinion network from the perspective of structural mechanics. This process specifically includes the following sub-steps: The system maps the semantic vector space of the execution nodes, and the time window is used. All valid nodes within the space are mapped to a high-dimensional semantic feature space. In practice, a pre-trained deep learning semantic embedding model is used to embed the content text of each node. Convert to a fixed-dimensional real vector The geometric position of the vector in the high-dimensional space directly represents the semantic features of the node viewpoint, and the cosine value of the angle between the vectors represents the semantic similarity between the viewpoints. For the choice of vector dimension, this embodiment selects 768-dimensional or 1024-dimensional vectors to ensure sufficient semantic expression capability.

[0047] To locate the mainstream viewpoint clusters and their centroids, the system performs density clustering analysis on the node vector set within the time window. In this embodiment, the DBSCAN algorithm is used to aggregate semantically similar nodes into clusters based on the cosine distance between nodes. The system counts the number of nodes in each cluster and the sum of the emotional potential of each node, defining the cluster with the largest total node potential as the mainstream viewpoint cluster at the current moment. .

[0048] Subsequently, the system calculates the centroid vector of the mainstream viewpoint cluster. To reflect the dominant role of high-potential nodes in the consensus direction, the centroid calculation adopts a weighted average method based on emotional potential energy, as shown in the following formula: ; in, Cluster node The emotional potential energy, For nodes The semantic vector, the centroid vector In the semantic space, it represents the mainstream consensus direction of the current public opinion field.

[0049] To filter the set of dissenting viewpoints, the system iterates through all nodes within the time window except for the mainstream viewpoint cluster, and calculates the vector for each node. With mainstream centroid vector Cosine similarity between them, setting a semantic similarity threshold. When the calculated cosine similarity is less than the threshold, the node is determined to be a dissenting node that significantly conflicts with the mainstream viewpoint. All dissenting nodes that meet the condition constitute the dissenting node set. .

[0050] Calculate semantic field topological stress This parameter is used to physically characterize the structural resistance intensity within the public opinion system. The system performs cumulative calculations on all nodes in the dissenting node set according to the following formula: ; The symbols in the formula are defined as follows: The system in time The total topological stress borne; the larger this value, the more severe the viewpoint tearing within the system. Dissent node The emotional potential energy indicates that only dissenting nodes with high potential energy can make a substantial stress contribution to the system structure; the stress contribution of noisy nodes with low potential energy, even if their views are opposed, can be ignored. Dissent node semantic vector; Cosine similarity between dissenting nodes and mainstream centroids, with a range of values. ; Semantic distance term: When dissenting opinions are completely opposed to mainstream opinions, this distance term takes the maximum value of 2; when the two are orthogonal and unrelated, this distance term takes the value of 1.

[0051] Through the above calculations, this invention transforms the abstract conflict of viewpoints into the specific physical quantity of topological stress. This physical quantity not only considers the semantic difference distance of viewpoints, but also couples the energy attributes of the viewpoint holders, thereby being able to distinguish between the noise of a minority and the destructive torrent of opposition. This feature provides a quantitative basis for subsequent determination of whether the system has reached the structural yield limit.

[0052] Meanwhile, in step S300, the system calculates the disorder index, i.e., system entropy, from the perspective of thermodynamic statistical physics based on the constructed weighted emotional potential energy field. This step aims to quantify the distribution of public opinion energy among network nodes, and is specifically implemented through the following sub-steps: To normalize the potential energy probability distribution and make it suitable for the information entropy calculation model, the system first needs to convert the physical-dimensional emotional potential energy into a dimensionless probability distribution. The system iterates through the current time window. Calculate the total effective emotional potential energy of all nodes within the node. Then, calculate each node. The probability of emotional potential energy : ; This probability value Satisfy the normalization condition, i.e. It physically represents a node. The relative share of energy it carries within the overall energy pool of the public opinion system.

[0053] Calculating the system entropy of the emotional potential field Based on the above probability distribution, the system uses the Shannon entropy formula to calculate the system entropy value at the current moment: ; The symbols are defined as follows: The system in time Information entropy, measured in bits; :node The potential energy probability distribution; Logarithmic operations with base 2.

[0054] System entropy The physical meaning of lies in characterizing the degree of orderliness of energy distribution within the public opinion system, when When the value is low, it indicates that the system's energy is highly concentrated in a few super nodes, and the public opinion field is in a steady-state structure of ordered consensus or unipolar dominance. At this time, the system has strong resistance to random disturbances. When the value increases significantly, it indicates that the system's energy is dispersed, with numerous small and medium-sized nodes simultaneously releasing high potential energy. The public opinion field is in a chaotic and disorderly state of cacophony. In this high-entropy state, the system lacks a dominant and stable core, making it highly susceptible to random drift or abrupt changes in its overall state under minor external stimuli. Real-time monitoring is crucial to address this. By observing the evolutionary trajectory of public opinion, this invention can capture the thermodynamic signs of the transition from order to disorder, which is usually the precursor to a reversal of public opinion.

[0055] After completing the calculation of the internal state parameters of the system, step S400 further establishes the dynamic response model of the system. In order to analyze the sensitivity of the public opinion system to external information, the system first needs to define and quantify the external stimulus input.

[0056] The system calculates the instantaneous external stimulus flow and identifies the current unit time step. The set of newly added nodes in the public opinion network These nodes represent external information energy newly entering the system. For each newly added node in the set... Calculate its effective emotional potential energy based on the formula in the aforementioned steps. .

[0057] Calculate cumulative external stimulus input To construct the horizontal axis of the hysteresis loop, the system performs time integration on the instantaneous stimulus flow to obtain the cumulative input within the observation period, calculated as follows: ; In a discrete-time system, this integration operation is transformed into an accumulation operation: ; The symbols are defined as follows: Deadline The cumulative total amount of external stimuli input. This physical quantity characterizes the total energy driving force continuously injected into the public opinion system from the outside world; The starting point of the observation period is usually set as the outbreak point of the public opinion event or the start point of monitoring. or At any moment or the The set of newly added nodes within a time step; Add a node The instantaneous and effective emotional potential energy.

[0058] This parameter This will be used as a driving variable in subsequent hysteresis loop models, by monitoring the system state as... The trajectory of the change can reveal the evolution law of the system after receiving external energy injection. For example, under normal steady state, the state response of the system should increase linearly or sublinearly with the increase of input; while in the reversal critical state, a small input increment may trigger a drastic change in the state response. This nonlinear relationship is the key basis for the present invention to identify reversal.

[0059] After calculating the cumulative external stimulus input as the driving variable, the system constructs an entropy-stress biphasic state space to characterize the internal stability response of the public opinion system.

[0060] The core of this step is to couple the previously independently calculated structural mechanical parameters with thermodynamic parameters through a multiphysics field to generate a unified dimensionless state modulus, which is used as the ordinate response variable of the hysteresis loop model.

[0061] Dimensionless normalization of execution parameters, due to topological stress With system entropy Because physical quantities and numerical orders of magnitude have different dimensions, direct spatial synthesis can cause features of large numerical dimensions to mask features of small numerical dimensions. Therefore, the system maintains a historical observation window and records the maximum stress value within that window. Minimum stress value Maximum entropy and minimum entropy value .

[0062] The system uses a range normalization method to map real-time parameters to... Interval: ; ; in, To normalize the topological stress, To normalize the system entropy, if the current real-time value exceeds the historical extreme value, the system will update the historical extreme value and recalculate the normalization, or truncate the overflow value to 1.0 or 0.0 to ensure the stability of the calculation.

[0063] Construct a two-phase state space and calculate the state modulus. The system establishes a two-dimensional Cartesian coordinate system to normalize the system entropy. The horizontal axis (X-axis) is used to represent the normalized topological stress. The vertical axis (Y-axis) is the coordinate system at any time. The system state is represented as a two-dimensional state vector. ; To compress the two-dimensional state into a one-dimensional response signal, the system calculates the weighted Euclidean norm of the state vector, defined as the biphase state modulus. : ; The symbols and parameters in the formula are defined as follows: The biphase state modulus is a physical quantity that comprehensively characterizes the degree to which a public opinion system deviates from its initial equilibrium state. The larger the value, the more intense the division of opinions and the chaotic energy dispersion that the system is simultaneously experiencing, meaning that the system is in an extremely unstable high-energy excited state.

[0064] Entropy weight coefficient: Used to adjust the contribution of system disorder to the total state modulus; its value range is... .

[0065] Stress weighting coefficient, used to adjust the contribution ratio of structural tension to the overall state modulus, with a value range of... .

[0066] In this embodiment, the default settings are... This means that structural fragmentation and disorder are considered to contribute equally to the reversal of public opinion. In other variant embodiments, the contribution can be appropriately increased if the topic is specific. The value is used to increase the sensitivity to force.

[0067] Through the above steps, this invention maps the complex process of public opinion evolution into... to The functional relationship, in normal public opinion dissemination, changes with the input... The increase of state modulus Typically, it exhibits linear or convergent growth; however, just before a reversal occurs, due to the nonlinear response of the system's internal structure, even a small input increment can lead to... The exponential increase in the input-response relationship forms the physical basis for subsequent hysteresis loop analysis.

[0068] After determining the driving variables With response variable Subsequently, the system constructs a dynamic hysteresis loop model and calculates its characteristic parameters. This process draws on the physical principle of hysteresis loops in ferromagnetic materials and uses the phase difference between the input and output quantities to characterize the social damping characteristics of the public opinion system.

[0069] Construct the input response phase trajectory, and the system operates within a specific observation period. Internally, a series of data point pairs are recorded in chronological order. ,in These discrete points are connected on a two-dimensional plane to form a continuous phase trajectory curve.

[0070] To form a closed loop for area calculation, the system needs to define loading and unloading processes. In the public opinion dynamics model of this invention, the loading process corresponds to the public opinion outbreak period, i.e., the input quantity. The rapid growth phase corresponds to the period of public opinion waning or plateauing, i.e., the phase where the input growth rate slows down significantly or returns to zero. If the actual data flow is monotonically increasing, the system adopts virtual closure technology, i.e., based on the current time point... As the turning point, a virtual regression curve is constructed connecting to the starting point of the cycle. This forms a closed loop. .

[0071] Calculate the closed area of ​​the hysteresis loop. The system uses Green's theorem or the polygon area formula to calculate the area of ​​the region enclosed by the aforementioned closed trajectory. In the discretization calculation, the trapezoidal integral method is used for approximate solution. ; The symbols are defined as follows: :time The area of ​​the hysteresis loop is a physical quantity that characterizes the ability of the public opinion system to dissipate energy in the current cycle, or the social damping of the system. The total number of sampling points within the observation period; : No. The coordinates of each sampling point.

[0072] In normal public opinion dissemination, social groups exhibit inertia in their acceptance and reaction to information, i.e., response variables. Changes will lag behind input variables The changes in [the area] create a hysteresis loop of a certain width, whose area [is related to the change in area]. Maintaining a relatively stable positive value range means that the system has good damping characteristics, can absorb external shocks, and maintain the stability of the pipe point structure.

[0073] Calculate the rate of change of the area of ​​the hysteresis loop The system performs differentiation on the continuously calculated area sequence to monitor the dynamic changing trend of the damping characteristics. ; in To monitor step length, when When the system exhibits stable fluctuations or positive growth, it indicates that the system damping is normal. When a significant negative value appears, it indicates that the hysteresis loop tends to flatten or even collapse. Physically, this means that the damping mechanism of the system fails, the inertia disappears, and small external input changes will instantly be transformed into a violent response of the system state. This damping failure phenomenon is the critical dynamic characteristic of the phase transition of the public opinion system.

[0074] In step S500, the system performs critical instability judgment for public opinion reversal. This process adopts dual escape criteria, that is, simultaneously satisfying the two-phase space yield judgment (condition one) and the damping failure judgment (condition two). First, for the two-phase space yield judgment, the judgment logic aims to identify whether the public opinion field has reached the stress limit for plastic deformation from the perspective of the static structural strength of the system.

[0075] By setting yield boundary conditions, a safe steady-state region is defined in the entropy-stress two-phase state space of the system. and yield instability region The specific boundary is determined by the following two threshold parameters: Stress yield limit This parameter characterizes the maximum degree of opinion fragmentation that the public opinion network structure can withstand. When the normalized topological stress... When this limit is exceeded, it means that the accumulated energy of dissenting opinions is sufficient to undermine the existing mainstream consensus structure. In this embodiment, Set to 0.85.

[0076] Critical entropy increase threshold This parameter characterizes the critical point at which a system transitions from order to disorder; when the normalized system entropy... When this threshold is exceeded, it means that the distribution of public opinion energy is extremely dispersed, and the system has lost its cohesion to maintain the status quo. In this embodiment, Set to 0.75.

[0077] The system performs real-time state vector region determination and monitors the current state vector in real time. If the state vector satisfies the following logical discriminant, then the system is determined to satisfy the two-phase space yield condition, i.e., condition one holds: ; The physical meaning of this condition is that the system has the structural basis for reversal only when it is simultaneously in a state of high stress tearing and high entropy disorder. If there is only high stress but low entropy, it belongs to a polarized steady state and is not easy to reverse. If there is only high entropy but low stress, it belongs to a noise dissipation state and will not trigger reversal either. Only when both conditions are met at the same time, that is, when there is both strong destructive force and lack of constraint force to maintain the structure, will the system yield and thus produce structural reversal. This judgment mechanism based on dual-parameter coupling effectively filters out false alarms that may be generated by a single indicator.

[0078] Following the aforementioned determination of static structural yielding, the system then proceeds to determine damping failure (condition two). This determination process focuses on the dynamic evolution characteristics of the public opinion system, aiming to identify whether the system has lost its buffering capacity against external information shocks. In a physical sense, a reversal of public opinion is often accompanied by the sudden disappearance of the system's social inertia, manifested as a sharp collapse of the hysteresis loop area.

[0079] To establish a historical reference for the hysteresis loop area, and to determine whether the system's damping state is abnormal at the current moment, it is necessary to obtain the system's average damping level under steady state. The system maintains a sliding historical window, and the moving average of the hysteresis loop area within this window is calculated. : ; in, This represents the total number of sampling points within the historical window. The average value is the sampling interval. This represents the system's ability to process external information in a timely manner and its level of inertia under normal conditions.

[0080] Calculate the damping attenuation coefficient The system will calculate the hysteresis loop area at the current moment. The current degree of damping contraction relative to normal is quantified by comparing it with historical benchmark values. The calculation formula is as follows: ; in, To prevent extremely small positive numbers with a denominator of zero; damping attenuation coefficient The numerical range is usually 100. ,when When this occurs, it indicates that the system damping is at a normal level, and the evolution of public opinion follows past inertia; when When the value is significantly less than 1, it indicates that the hysteresis loop is contracting and the phase lag effect of the system is weakening.

[0081] Threshold discrimination for damping failure; the system sets an area shrinkage threshold. This threshold defines the critical point at which the system loses steady-state inertia, and is combined with the calculated rate of change of area. The system determines whether condition two is true based on the following logical discriminant: ; The physical meaning of this discriminant is extremely clear: the first term This indicates that the hysteresis loop has collapsed into an extremely narrow, elongated loop or straight line. At this point, the system input... With response There is almost no phase difference between them, meaning that the public's acceptance of new information no longer goes through a period of rational digestion or questioning, but instead exhibits an immediate, synchronous, and excessive reaction. (The second item...) This ensures that the area is in a dynamic process of continuous shrinkage, eliminating the possibility of misjudging the low area due to long-term silence.

[0082] When the area of ​​the hysteresis loop shrinks sharply to below the critical value, it is physically determined that the social damping of the system has failed. At this time, the system loses its ability to absorb external energy oscillations. Any small external negative stimulus will no longer be attenuated by the inertia of the system, but will be directly, without loss, or even amplified into a drastic change in the state of the system. This sudden change in dynamic characteristics is the most typical physical sign before the occurrence of a public opinion reversal event, and constitutes an indispensable dynamic criterion in the prediction logic of this invention.

[0083] After completing the two-phase space yield test (condition one) and the damping failure test (condition two) respectively, the system performs comprehensive logical operations on multi-dimensional parameters to generate the final public opinion reversal early warning signal. This part aims to transform the discrete physical field parameters into standardized data instructions that can be recognized by the decision-making system.

[0084] The system performs a double escape logic AND operation and reads the static yield determination result. and the output dynamic damping failure judgment results The public opinion reversal advocated in this invention is strictly defined as the simultaneous instability of the system in both structural strength and dynamic inertia dimensions. Therefore, the system generates a reversal trigger flag based on the following Boolean logic. : ; If and only if True (i.e., the system is in a high-stress, high-entropy state) and When true (i.e., the hysteresis loop collapses and the damping fails), The strict AND logic, set to 1, forms the core filtering mechanism of this invention, effectively eliminating false signals from a single dimension: for example, it eliminates normal hot topics that are hotly debated but structurally stable, as well as short-lived disturbances that are rapid in response but lack destructive power.

[0085] Calculate the probability index of public opinion reversal risk In addition to binary trigger flags, to provide finer-grained early warning, the system constructs a continuous risk probability model based on the Sigmoid activation function. The system first calculates the weighted deviation of each key parameter from its safety threshold. : ; in, , , These are weighting coefficients, corresponding to the contribution weights of topological stress, system entropy, and damping coefficient, respectively. In this embodiment, they are set as follows: It should be noted that for the damping term, since a smaller value indicates a greater risk, a [specific method / approach] is used. The form ensures the degree of deviation. Positively correlated with risk.

[0086] Then, the normalized risk probability index is calculated: ; in, This is the slope adjustment parameter, used to control the steepness of the probability curve; The range of values ​​is The closer this value is to 1, the higher the certainty of a reversal in the public opinion system. This calculation process maps multi-source heterogeneous physical parameters to a unified probability space, enabling the system to quantify and output a specific value of the reversal probability, rather than just a simple alarm.

[0087] The system generates and distributes early warning data packets based on calculations. and Construct a standardized JSON data packet containing the following fields: Timestamp: Calculates the precise time of occurrence; State modulus: the current state modulus. Numerical value; Risk probability: the calculated ; Core attribution: In calculation The sub-item that contributes the most at the time is used to assist manual judgment; Warning level: According to The interval (e.g.) They are marked as safe, concerned, and high risk, respectively.

[0088] The system pushes the data packet to the downstream visualization terminal or decision support system in real time through a pre-built RESTful API interface or message queue. The serialization transmission of the data packet and the network communication protocol are common technical means in the field of computer communication and will not be described in detail here. In this way, the present invention transforms the abstract physical field calculation results into intuitive digital instructions to support subsequent public opinion intervention operations.

Claims

1. A method for identifying public opinion reversals based on dynamic changes in emotional energy, comprising the following steps: Acquire online public opinion data within a set time window, map network nodes to semantic vectors, and calculate the effective emotional potential of each node; Based on the weighted emotional potential energy field, the semantic field topological stress, which characterizes the degree of tearing in the viewpoint structure, and the emotional potential energy field system entropy, which characterizes the degree of disorder in energy distribution, are calculated respectively. The cumulative external stimulus input of online public opinion is calculated, and the semantic field topological stress and the emotional potential field system entropy are coupled into a biphasic state modulus. A hysteresis loop is generated based on the dynamic relationship between the cumulative external stimulus input and the biphasic state modulus. Monitor the evolution characteristics of the biphase state modulus and the hysteresis loop, and output a public opinion reversal early warning signal when the dual escape discrimination is satisfied; The dual escape criterion includes simultaneously satisfying the two-phase space yield criterion and the damping failure criterion.

2. The method for identifying public opinion reversals based on dynamic changes in emotional energy according to claim 1, characterized in that, The calculation of the effective emotional potential energy of each node includes: The text sentiment intensity modulus, propagation centrality weight, and local semantic redundancy of each node are obtained respectively. The product of the emotional intensity modulus and the propagation centrality weight is used as the gain term; The sum of the local semantic redundancy and the preset smoothing coefficient is used as the damping term; The effective emotional potential energy of the node is obtained by calculating the ratio of the gain term to the damping term.

3. The method for identifying public opinion reversals based on dynamic changes in emotional energy according to claim 1, characterized in that, The semantic field topological stress, which characterizes the degree of tearing in the viewpoint structure, includes: Identify the mainstream viewpoint clusters in the semantic vector space and calculate the centroid vector of the mainstream viewpoint clusters: Identify dissenting nodes that do not belong to the mainstream view cluster, and calculate the relative position vector between the semantic vector of the dissenting node and the centroid vector; The relative position vector is weighted using the effective emotional potential energy of dissenting nodes, and the equivalent stretching force of dissenting nodes on the mainstream viewpoint cluster is calculated. The modulus of the force is determined as the topological stress of the semantic field.

4. The method for identifying public opinion reversals based on dynamic changes in emotional energy according to claim 1, characterized in that, The calculation of the entropy of the emotional potential energy field system, which characterizes the degree of disorder in energy distribution, includes: The sum of the effective emotional potential energy of all nodes within the current time window is calculated. Calculate the proportion of the effective emotional potential energy of a single node to the total, and use it as the probability of the node's potential energy proportion. Based on the potential energy proportion probability of all nodes, the entropy of the emotional potential energy field system is calculated using the Shannon entropy definition logic.

5. The method for identifying public opinion reversals based on dynamic changes in emotional energy according to claim 1, characterized in that, The calculation of cumulative external stimulus input and biphasic state modulus in the construction of the dynamic response model includes: Identify the set of nodes that have newly entered the network within the current time step, and perform time integration or summation on the effective emotional potential of all nodes in the set to obtain the cumulative external stimulus input; The semantic field topological stress and the emotional potential field system entropy are normalized respectively. A two-dimensional state space is constructed with the normalized system entropy as the horizontal axis and the normalized topological stress as the vertical axis. The weighted Euclidean norm of the current state vector in the two-dimensional state space is calculated to obtain the two-phase state modulus.

6. The method for identifying public opinion reversals based on dynamic changes in emotional energy according to claim 1, characterized in that, The generation of hysteresis loops based on the dynamic relationship between the accumulated external stimulus input and the biphasic state modulus includes: Using the accumulated external stimulus input as the driving variable and the biphasic state modulus as the response variable, an input-response phase trajectory is constructed. Within the observation period, the beginning and end of the phase trajectory are connected to form a closed loop, and the area enclosed by the closed loop is calculated using a numerical integration method. The area of ​​the region is defined as the area of ​​the hysteresis loop, which characterizes the social damping properties of the public opinion system.

7. The method for identifying public opinion reversals based on dynamic changes in emotional energy according to claim 1, characterized in that, The specific yield determination condition for the two-phase space is as follows: Preset stress yield limit and critical entropy increase threshold; Determine whether the current normalized topological stress is greater than the stress yield limit; Simultaneously determine whether the current normalized system entropy is greater than the critical entropy increase threshold; If both of the above judgment results are yes, then the two-phase space yield judgment condition is satisfied.

8. The method for identifying public opinion reversals based on dynamic changes in emotional energy according to claim 6, characterized in that, The damping failure determination criteria are as follows: Calculate the moving average of the hysteresis loop area within the historical sliding window as a benchmark reference value; Calculate the ratio of the hysteresis loop area at the current moment to the reference value to obtain the damping attenuation coefficient; Calculate the rate of change of the area of ​​the hysteresis loop over time; If the damping attenuation coefficient is less than the preset area shrinkage threshold and the rate of change is negative, then the damping failure determination condition is met.

9. The method for identifying public opinion reversals based on dynamic changes in emotional energy according to claim 1, characterized in that, The output of the public opinion reversal early warning signal includes: Perform a Boolean AND operation on the results of the two-phase space yield test condition and the damping failure test condition. A reversal warning data packet containing the current biphase state modulus value is generated only if the calculation result is true.

10. A method for identifying public opinion reversals based on dynamic changes in emotional energy, as described in claim 8, is characterized in that... The method also includes calculating a probability index of public opinion reversal risk: Calculate the deviation of the normalized topological stress, normalized system entropy, and damping attenuation coefficient from their respective preset thresholds. The total deviation is obtained by weighted summation of the various deviations. The total deviation is mapped to a value between 0 and 1 using the Sigmoid function, which serves as the probability index of public opinion reversal risk and is output along with the public opinion reversal warning signal.

Images